1. Field of the Invention
The present invention relates to an image processing technique.
2. Description of the Related Art
An image search apparatus which records representative images for a moving image is disclosed in, for example, Japanese Patent Laid-Open No. 2000-222584 (to be referred to as patent reference 1 hereinafter). With the technique disclosed in patent reference 1, a specific object is detected from each frame which forms a moving image, and a feature amount associated with the detected object, that associated with the background, and the difference between the object and background are recorded outside the moving image.
A method of extracting a frame including a predetermined object image as a representative frame is disclosed in, for example, Japanese Patent Laid-Open No. 10-224736 (to be referred to as patent reference 2 hereinafter). With the technique disclosed in patent reference 2, whether or not a partial region in an image includes a specific object is evaluated using a dictionary. Then, a frame which is evaluated to be closer to that registered in the dictionary is extracted as a representative frame.
However, since the technique disclosed in patent reference 1 detects a specific object using a motion vector, when a motionless region in an image includes the specific object, that object cannot be detected. For this reason, although the redundancy of the entire image is reduced, the contents of a moving image itself cannot be recognized. Therefore, it is difficult to play back or edit the moving image for an optimal time according to the object contents.
With the technique disclosed in patent reference 2, since only a given area is evaluated, representative frames may often include similar image patterns. In this case, even when such moving image period is searched, a large number of representative frames having nearly the same contents are generated, and this technique is not suited to recognize the contents in a moving image. For this reason, it is not easy to play back or edit the moving image from an optimal time according to the object contents.